CA2441544A1

CA2441544A1 - System, method and apparatus that isolate virtual private network (vpn) and best effort traffic to resist denial of service attacks

Info

Publication number: CA2441544A1
Application number: CA002441544A
Authority: CA
Inventors: David E. Mcdysan
Original assignee: Worldcom Inc
Current assignee: Verizon Business Global LLC
Priority date: 2001-03-20
Filing date: 2002-03-20
Publication date: 2002-09-26
Also published as: EP1371178A1; JP2004528756A; WO2002076029A1; MXPA03008476A; CN1500329A; EP1371178A4; BR0208227A

Abstract

A network architecture (20) in accordance with the present invention includes a communication network that supports one or more network-based Virtual Private Networks (VPNs). The communication network includes a plurality of boundary routers (22a-22d) that are connected by access links to CPE edge routers (24b-24d and 25a-25d) belonging to the one or more VPNs. To prevent traffic from outside a customer's VPN (e.g., traffic from other VPNs or the Internet at large) from degrading the QoS provided to traffic from within the customer's VPN, the present invention gives precedence to intra-VPN traffic over extra-VPN traffic on each customer's access link through access link prioritisation or access link capacity allocation, such that extra-VPN traffic can not interfere with inter-VPN traffic.

Description

SYSTEM, METHOD AND APPARATUS THAT ISOLATE VIRTUAL PRIVATE
NETWORK (VPN) AND BEST EFFORT TRAFFIC TO RESIST DENIAL OF
SERVICE ATTACKS
The present invention relates to communication networks and, in particular, to the prevention of denial of service attacks in a public communication network, for example, the Internet. Still more particularly, the present invention relates to method, system and apparatus for preventing denial of service attacks in a communication network having a shared network infrastructure by separating the allocation and/or prioritization of access capacity to traffic of sites within a virtual private network (VPN) from the allocation andlor prioritization of access capacity to sites in another VPN or the public network.
For network service providers, a key consideration in network design and management is the appropriate ahocation of access capacity and network resources between traffic originating from VPN customer sites and traffic originating from outside the VPN (e.g., from the Internet or other VPNs). This consideration is particularly significant with respect to the traffic of VPN
customers whose subscription includes a Service Level Agreement (SLA) requiring the network service provider to provide a minimum communication bandwidth or to guarantee a particular Quality of Service (QoS). Such service offerings require the network service provider to implement a network architecture and protocol that achieve a specified QoS and ensure sufficient access capacity and network resources are available for communication with,other VPN sites separate from communication with hosts that are not part of the VPN.
In Internet Protocol (IP) networks, a straightforward approach to achieving QoS and implementing admission control comparable to that of connection-oriented network services, such as voice or Asynchronous Transfer Mode (ATM), is to emulate the same hop-by-hop switching paradigm of signaling resource.reservations for the flow of IP
packets requiring QoS.
In fact, the IP signaling standard developed by the Internet Engineering Task Force (IETF) for Integrated Services (Intserv) adopts precisely this approach. As described in IETF RFC 1633 [R.
Branderi et al., "Integrated Services in the Internet Architecture: an Overview" June 1994], Intserv is a per-flow IP QoS architecture that enables applications to choose among multiple, controlled levels of delivery service for their data packets. To support this capability, Intserv permits an application at a transmitter of a packet flow to use the well-known Resource ReSerVation Protocol (RSVP) defined by IETF RFC 2205 [R. Branden et al., "Resource ReSerVation Protocol (RSVP) - Version 1 Functional Specification" Sept. 1997]
to request a desired QoS class at a specific level of capacity from all network elements along the path to a receiver of the packet flow. After receiving an RSVP PATH message requesting a resource reservation and an RSVP RESV message confirming resource reservation from an upstream node, individual network elements along the path implement mechanisms to control the QoS and capacity delivered to packets within the flow.
Figure 1 illustrates the implications ofutilizing a conventional Intserv implementation to perform admission control. As shown in Figure 1, an exemplary IP network 10 includes N
identical nodes (e.g., service provider boundary routers) 12, each having L
links of capacity X
coupled to Customer Premises Jquipment (CPE) 14 for ~L distinct customers. In a per-flow, connection-oriented approach, each node 12 ensures that no link along a network path from source to destination is overloaded. Looking at access capacity, a per-flow approach is able to straightforwardly limit the input flows on each of the ingress access links such that the sum of the capacity for all flows does not exceed the capacity X of any egress access link (e.g., Link 1 of node 12a). A similar approach is applicable to links connecting unillustrated core routers within IP network 10. ' Although conceptually very simple, the admission control technique illustrated in Figure 1 has a number of drawbacks. Most importantly, Intserv admission control utilizing RSVP has limited scalability because of the processing-intensive signaling RSVP
requires in the service provider's boundary and core routers. In particular, RSVP requires end-to-end signaling to request appropriate resource allocation at each network element between the transmitter and receiver, policy queries by ingress node 12b-12d to determine which flows to admit and police their traffic accordingly, as well as numerous other handshake messages.
Consequently, the processing required by Intserv RSVP signaling is comparable to that of telephone or ATM
signaling and requires a high performance (i.e., expensive) processor component within each boundary or core IP router to handle the extensive processing required by such signaling. RSVP

signaling is soft state, which means the signaling process is frequently refreshed (by default once every 30 seconds) since the forwarding path across the IP network may change and therefore information about the QoS and capacity requested by a flow must be communicated periodically.
This so-called soft-state mode of operation creates an additional processing load on a router even greater than that of an ATM switch. Furthermore, if the processor of a boundary router is overloaded by a large number of invalid RSVP requests, the processor may crash, thereby disrupting service for all flows for all customers being handled by the router with the failing processor.
In recognition of the problems associated with implementing admission control utilizing conventional Intserv RSVP signaling, the IETF promulgated the Differentiated Services (Diffserv or DS) protocol defined in RFC 2475 [S. Blake. et al., "An Architecture for Differentiated Services" Dec. 1998]. Diffserv is an IP QoS architecture that achieves scalability by conveying an aggregate traffic classification within a DS field (e.g., the IPv4 Type of Service (TOS) byte or IPv6 traffic class byte) of each IP-layer packet header. The first, six bits of the DS
field encode a Diffserv Code Point (DSCP) that requests a specific class of service or Per Hop Behavior (PHB) for the packet at each node along its path within a Diffserv domain.
In a Diffserv domain, network resources are allocated to aggregates of packet flows in.
accordance with service provisioning policies, which govern DSCP marking and traffic conditioning upon entry to the Diffserv domain and traffic forwarding within the Diffserv domain. The marking (i.e., classification) and conditioning operations need be implemented only at Diffserv network boundaries. Thus, rather than requiring end-to-end signaling between the transmitter and receiver to establish a flow having a specified QoS, Diffserv enables an ingress boundary muter to provide the QoS to aggregated flows simply by examining and/or marking each IP packet's header.
Although the Diffserv standard addresses Intserv scalability limitation by replacing Intserv's processing-intensive signaling with a simple per packet marking operation that can easily be performed in hardware, implementation of the Diffserv protocol presents a different type of problem. In particular, because Diffserv allows host marking of the service class, a Diffserv network customer lime can experience a Denial of Service (DoS) attack if a number of _ Q, _ hosts send packets to that link with the DS field set to a high priority. It should be noted that a set of hosts can exceed the subscribed capacity of a Diffserv service class directly by setting the DSCP or indirectly by submitting traffic that is classified by some other router or device to a particular DSCP. In Diffserv, an IP network can only protect its resources by policing at the ingress routers to ensure that each customer interface does not exceed the subscribed capacity for each Diffserv service class. However, this does not prevent a DoS attack.
Figure 2 depicts a DOS attack scenario in an exemplary IP network 10' that implements the conventional Diffserv protocol. In Figure 2, a number of ingress nodes (e.g., ingress boundary routers) 12b'-12d' each admit traffic targeting a single link of an egress node (e.g., egress boundary router) 12a'. Although each ingress nodes 12' polices incoming packets to ensure that customers do not exceed their subscribed resources at each DSCP, the aggregate of the admitted flows exceeds the capacity X of egress Link 1 of node 12a', resulting in a denial of service,to the customer site served by this link.
In view of the limitations attendant to conventional implementations of the Intserv and Diffserv standards, the present invention recognizes that it would be useful and desirable to provide a method, system and apparatus for data communication that support a communication protocol that, unlike conventional Intserv implementations, is highly scalable and yet protects against the DoS attacks to which conventional Diffserv and other networks are susceptible.
A network architecture in accordance with the present invention includes a communication network that supports one or more network-based Virtual Private Networks (VPNs). The communication network includes a plurality of boundary routers that are connected by access links to CPE edge routers belonging to the one or more VPNs. To prevent traffic from outside a customer's VPN (e.g., traffic from other VPNs or the Internet at large) from degrading the QoS provided to traffic from within the customer's VPN, the present invention gives precedence to intra-VPN traffic over extra-VPN traffic on each customer's access link through access link prioritization or access link capacity allocation, such that extra-VPN traffic cannot interfere with inter-VPN traffic. Granting precedence to intra-VPN traffic over extra-VPN
traffic in this manner entails special configuration of network elements and protocols, including partitioning between intra-VPN and extra-VPN traffic on the physical access link and access network using layer 2 switching and multiplexing, as well as the configuration of routing protocols to achieve logical traffic separation between intra-VPN traffic and extra-VPN traffic at the VPN boundary routers and CPE edge routers. By configuring the access networks, the VPN
boundary routers and CPE edge routers, and the routing protocols of the edge and boundary routers in this manner, the high-level service of DoS attack prevention is achieved.
Additional objects, features, and advantages of the present invention will become apparent from the following detailed written description.
The novel features believed characteristic of the invention are set forth in the appended claims. The invention itself however, as well as a preferred mode of use, further objects and advantages thereof, will best be understood by reference to the following detailed description of an illustrative embodiment when read in conjunction with the accompanying drawings, wherein:
Figure 1 depicts a conventional Integrated Services (Intserv) netwoxk that implements per-flow QoS utilizing RSVP;
Figure 2 illustrates a conventional Differentiated Services (Diffserv) network that implements QoS on aggregated traffic flows utilizing DSCP markings in each packet header and is therefore vulnerable to a Denial of Service (DoS) attack;
Figure 3 depicts an exemplary communication network that, in accordance with a preferred embodiment of the present invention, resists DoS attacks by partitioning allocation and/or prioritization of access capacity by reference to membership in Virtual Private Networks (VPNs);
Figure 4 illustrates an exemplary network architecture that provides a CPE-based VPN
solution to the DoS attack problem;
Figure 5 is a more detailed block diagram of a QoS-aware CPE edge router that maybe utilized within the network architectures depicted in Figures 4 and 7;
Figure 6A is a more detailed block diagram of a QoS-aware boundary router without VPN function that may be utilized within the network architectures illustrated in Figures 4 and 7;
Figure 613 is a more detailed block diagram of a QoS-aware boundary router having s VPN function that may be utilized within the network architecture illustrated in Figure 4;
Figure 7 illustrates an exemplary network architecture that provides a network-based VPN solution to the DoS attack problem; and Figure 8 is a more detailed block diagram of a QoS-aware VPN boundary router that may be utilized within the network architecture depicted in Figure 7.
With reference again to the figures and, in particular, with reference to Figure 3, there is depicted a high level block diagram of an exemplary network architecture 20 that, in accordance with the present invention, provides a scalable method of providing QoS to selected traffic while protecting a Virtual Private Network (VPN) customer's access and trunk network links against DoS attacks. Similar to the prior art network illustrated in Figure 2, network architecture 20 of Figure 3 includes a Diffserv network 21 having N service provider boundary routers (BRs) 22 that each have L access links. What is different in network architecture 20 is that Diffserv network 21 supports a plurality of VPN instances, of which two are shown in the figure as identified by the access links of boundary routers 22 coupled to CPE edge routers (ERs) for a first network service customer 24 and an ER for a second network service customer 25 at each of four sites, respectively identified by letters a through d. Each CPE ER
provides network service to a customer's local area networks (LANs). The service provider network-based VPN may support many more customers than the two shown in this figure.
In the exemplary communication scenario depicted in Figure 3, hosts within the LANs of the first VPN customer coupled to CPE edge routers 24b-24d, those within a second VPN
customer's LANs coupled to CPE edge routers 25a-25d, as well as sites coupled to other unillustrated CPE edge routers linked to boundary routers 22a-22d, may all transmit packet flows targeting the LAN coupled to the first VPN customer CPE edge router 24a.
If the conventional Diffserv network of the prior art described above with respect to Figure 2 were implemented, the outgoing access link 1 of boundary router 22a coupled to CPE
edge router 24a could be easily overwhelmed by the convergence of these flows, resulting in a DoS. However, in accordance with the present invention, Diffserv network 21 of Figure 3 prevents a DoS attack from sites outside the VPN by directing intra-VPN traffic to a first logical port 27 on physical access link 1 of boundary router 22a, while directing traffic from other VPNs or other sites to a second logical port 28 on physical access link 1 of boundary router 22a.
To prevent traffic from outside a customer's community of interest (e.g., traffic from other VPNs or the Internet at large) from degrading the QoS provided to traffic from within the customer's community of interest (e.g., traffic from other hosts in the same business enterprise), the present invention either prioritizes infra-VPN traffic over extra-VPN
traffic, or allocates access link capacity such that extra-VPN traffic cannot interfere with inter-VPN traffic. In other words, as described in detail below, each boundary router 22 gives precedence on each customer's access link to traffic originating within the customer's VPN, where a VPN is defined herein as a collection of nodes coupled by a shared network infrastructure in which network resources and/or communications are partitioned based upon membership of a collection of nodes. Granting precedence to infra-VPN traffic over extra-VPN traffic in this manner entails special configuration of network elements and protocols, including partitioning of the physical access between infra-VPN and extra-VPN traffic using layer 2 multiplexing and the configuration of routing protocols to achieve logical traffic separation. In summary, the configuration of the CPE edge router, the access network, the network-based VPN boundary router and the routing protocols involved in the edge and boundary routers cooperate to achieve the high-level service of DoS attack prevention, as detailed below.
Conventional Diffserv and .
CPE edger router IPsec-based IP VPN implementations, by contrast, do not segregate traffic destined for sites within the same VPN (i.e., infra-VPN traffic) and traffic sent from other regions of the Internet (i.e., extra-VPN traffic).
Referring now to Figures 4-8, at least two classes of implementations of the generalized network architecture 20 depicted in Figure 3 are possible. In particular, a network in accordance with the present invention can be realized as a CPE-based VPN implementation, as described below with reference to Figures 4-6, or as a network-based VPN implementation, as described .
below with reference to Figures 7-8.
Referring first to Figure 4, there is illustrated an exemplary network architecture 30 that employs a CPE-based VPN to resist DoS attacks. The depicted network architecture includes a.
Diffserv-enabled IP VPN network 44, a best effort TP public network 46, and a plurality of customer Local Area Networks (LANs) 32. Customer LANs 32 each include one or more hosts - g-48 that can function as a transmitter and/or receiver of packets communicated over one or both of networks 44 and 46. In the exemplary implementation illustrated in Figure 4, it is assumed that customer LANs 32a and 32b belong to the same community of interest (i.e.,VPN), such as a business enterprise. , Each customer LAN 32 is coupled by a respective CPE edge router 34 and physical access link 35 to a respective access network (e.g., an L2 access network) 38.
Access networks 38a and 38b each have a first L2 access logical connection to a boundary router (BR) 40 of Diffserv-enabled IP VPN network 44 and a second L2 access logical connection to a boundary router (BR) 42 of best effort IP public network 46. As illustrated in Figure 4 by differing line styles representing infra-VPN and extra-VPN traffic, VPN-aware CPE edge routers 34a and 34b route only packets with IP address prefixes belonging to the If VPN via Diffserv-enabled IP
VPN network 44, arid route all other traffic via best effort IP public network 46. To enhance security of customer LANs 32, CPE edge routers 34a and 34b send all traffic to and from best effort IP public network 46 through a respective one of firewalls 36a and 36b.
In the network architecture illustrated in Figure 4, DoS attacks originating outside of the IP VPN are prevented by configuration of boundary routers 40a-40b and 42a-42b to appropriately utilize the two logical connections of access networks 38a and 38b to grant precedence to infra-VPN traffic. For example, in a first configuration, a higher priority is assigned to the L2 access logical connection with Diffserv-enabled IP VPN
network 44 than to the L2 access logical connection with best effort public IP network 46. L2 access networks that support~such prioritization of access links 35 include Ethernet (e.g., utilizing Ethernet priority), ATM (e.g., utilizing ATM service categories), and many frame relay (FR) network implementations. These implementations can each be provisioned utilizing well-known techniques. With this configuration, each boundary router 40 of Diffserv enabled IP VPN
network 44 shapes the transmission rate of packets to its logical connection to access network 38 to a value less than that of the access link to prevent starvation of the L2 access logical connection to best effort IP public network 46. Alternatively, in a second configuration, boundary routers 40a-40b and 42a-42b may be individually configured to shape the traffic destined for each L2 access network logical connection to a specified rate, where the sum of s these rates is less than or equal to the transmission capacity of the physical access medium linking CPE edge routers 34 and access networks 38. In either of these alternative configurations, boundary routers 40 and 42 perform scheduling and prioritization based upon packets' DSCP markings and shape to the capacity allocated to the access network connection for IP VPN traffic.
As will be appreciated by those skilled in the art, selection of which of the alternative configurations to implement is a matter of design choice, as each configuration has both advantages and disadvantages. For example, with the first configuration, coordination of the access network configuration between networks 44 and 46 is easier. However, if access networks 38 implement only strict priority, then IP VPN traffic from Diffserv-enabled IP VPN
network 44 may starve best effort traffic communicated over IP public network 46. The second configuration addresses this disadvantage by allocating a portion of the access link capacity to each type ofnetwork access (i.e., both intra-VPN and extra-VPN). However, ifboundaryrouters 40 and 42 shape traffic in accordance with the second configuration, unused access capacity to one of networks 44 and 46 cannot be used to access the other network. That is, since the shapers are on separate boundary routers 40 and 42, only non-work-conserving scheduling is possible.
With reference now to Figure 5, there is illustrated a more detailed block diagram of a QoS-aware CPE edge router 34 that may be utilized within the network architecture depicted in Figure 4. As illustrated, CPE edge router 34 includes a number of LAN ports 60, which provide connections for a corresponding number of customer LANs 32. For example, in Figure 5, LAN
port 60a is connected to a customer LAN 32 including a number of hosts 48 respectively assigned 32-bit IP addresses "a.b.c.d," "a.b.c.e.," and "a.b.c.f."
Each LAN port is also coupled to a forwarding function 62, which forwards packets between LAN ports 60 and one or more logical ports (LPs) 66 residing on one or more Wide Area Network (WAN) physical ports 64 (only one of which is illustrated). LPs 66, which each comprise a layer-2 sub-interface, may be implemented, for example, as an Ethernet Virtual LAN
(VLAN), FR Data Link Connection Identifier (DLCI), ATM Virtual Channel Connection (VCC), or Point-to-Point Protocol (PPP)/ High-Level Data Link Control (HDLG) running on a Time Division Multiplexed (TDM) channel. WAN physical port 64 employs a scheduler 68 to g.

multiplex packets from logical ports 64 onto the transmission medium of an access network 38 and forwards packets received from access network 38 to the respective logical port utilizing a forwarding function 70.
When a LAN port 60 of CPE edge router 34 receives packets from a customer LAN
32, the packets first pass through a classifier 80, which determines by reference to a classifier table 82 how each packet will be handled by CPE edge router 34. As illustrated in Figure 5, classifier table 82 may have a number of indices, including Source Address (SA) and Destination Address (DA), Source Port (SP) and Destination Port (DP), Protocol Type (PT), DSCP, or other fields from packets' link, network or transport layer headers. Based upon a packet's values for one or more of these indices, classifier 72 obtains values for a politer (P), marker (M), destination LP, and destination LP queue (Q) within CPE edge router 34 that will be utilized to process the packet. In alternative embodiments of the present invention, lookup of the destination-LP and destination LP queue entries could be performed by forwarding function 62 rather than classifier 80.
As shown, table entry values within classifier table 82 may be fully specified, partially specified utilizing a prefix or range, or null (indicated by "-"). For example, the SAs of hosts 48 of LAN 32 are fully specified utilizing 32-bit IP addresses, DAs of several destination hosts are specified utilizing 24-bit IP address prefixes that identifyparticular IP
networks, and a number of index values and one policing value are null. In general, the same politer, marker, and/or shaper values, which for Intserv flows are taken from RSVP RESV messages, may be specified for different classified packet flows. For example, classifier table 82 specifies that politer P1 and marker M1 will process packets from any SA marked with DSCP "101" as well as packets having a SA "a.b.c.e" marked with DSCP "010." However, classifier table 82 distinguishes between flows having different classifications by specifying different destination LP values for traffic having a DA within the VPN (i.e., intra-VPN traffic) and traffic addressed to hosts elsewhere in the Internet (i.e., extra-VPN traffic). Thus, because IP address prefixes "r.s.t,"
"w.x.y," and "l.m.n" all belong to the same VPN as network 32, traffic matching these DAs is sent via LP-1 66a to other sites within the same VPN over the Diffserv-enabled IP VPN network 44 while all other traffic is sent via LP-2 66b to best effort IP public network 46.
10.

The logical port 66 and LP queue to which packets are forwarded can be determined by static configuration or dynamically by a, routing protocol. . In either case, a VPN route should always have precedence over an Internet route if a CPE router 34 has both routes installed for the same destination IP address. Such priority can be achieved in any of several ways, including (1) use of Interior Gateway Protocol (IGP) (i.e., OSPF and IS-IS) to install VPN
routes and EBGP
or static routing to install Internet routes or (2) use of EBGP to install both VPN routes and Internet routes, with a higher local preference being given for VPN routes.
After classification, packets are policed and marked, as appropriate, by policers P0, P 1 and markers M0, Ml, M2 as indicated by classifier table 82 and then switched by forwarding function 62 to either logical port 66a or 66b, as specified by the table lookup. Within the specified logical port 66, packets are directed to the LP queues QO-Q02 specified by classifier table 82. LP queues QO-Q2 perform admission control based upon either available buffer capacity or thresholds, such as Random Early Detection (RED). A scheduler 90 then services LP
queues QO-Q2 according to a selected scheduling algorithm, such as First In, First Out (FIFO), Priority, Weighted Round Robin (WRR), Weighted Fair Queuing . (WFQ) or Class-Based Queuing (CBQ). For example, in the illustrated embodiment, scheduler 90 of LP-2 66a implements WFQ based upon the weight w; associated with each LP queue i and the overall WFQ scheduler rate r2 for logical port 2, thereby shaping traffic to the rate ra. Finally, as noted above, scheduler 68 of physical WAN port 64 services the various logical ports 66 to control the transmission rate to access network 38.
CPE edge router 34 receives packets from access network 38 at WAN physical port 64 and then, utilizing forwarding function 70, forwards packets to the appropriate logical port 66a or 66b as indicated by configuration of access network 38 as it maps to the logical ports. At each logical port 66, packets pass through a classifier 100, which generally employs one ormore indices within the same set of indices discussed above to access a classifier table 102. In a typical implementation, the lookup results of classifiers 100 are less complex than those of classifier 80 because policing and marking are infrequently required. Thus, in the depicted embodiment, packets are forwarded by forwarding function 62 directly from classifiers 100 of logical ports 66 to the particular queues QO-Q2 of LAN port 60a specified in the table lookup m -1~
based upon the packets' DSCPs. As described above, queues QO-Q2 of LAN port 60a are serviced by a scheduler 102 that implements WFQ and transmits packets to customer LAN 32.
Referring now to Figure 6A, there is depicted a more detailed block diagram of a QoS-aware boundary router without any VPN function, which may be utilized within the network architecture of Figure 4, for example, to implement boundary routers 42. As shown, boundary router 42 of Figure 6A includes a plurality of physical ports 116, a plurality of logical ports 110 coupled to access network 38 by a forwarding function 112 for incoming packets and a scheduler 114 for outgoing packets, and a forwarding function 118 that forwards packets between logical ports 110 and physical ports 116. The implementation of multiple physical ports 116 permits fault tolerant connection to network core routers, and the implementation of multiple logical ports coupled to access network 38 permits configuration of one logical port (i.e., LP-1 110x) as a Diffserv-enabled logical port and a second logical port (i.e., LP-Z 110b) as a best-effort logical port. .
Thus, for traffic communicated from access network 38 through LP-2110b of boundary router 42 towards the network core, classifier 124 of LP-2110b directs all packets to marker MO
in accordance with classifier table 126. Marker MO remarks all packets received at LP-2110b with DSCP 000, thus identifying the packets as best-effort traffic. Classifier 120 of LP-1 110a, by contrast, utilizes classifier table 122 to map incoming packets, which have already received DSCP marking at a trusted CPE (e.g., service provider-managed CPE edge router 34), into queues QO-Q2 on PHY-I 116a, which queues are each associated with a different level of QoS.
Because the packets have already been multi-field classified, marked and shaped by the trusted CPE, boundary router 42 need not remark the packets. If, however, the sending CPE edge router were not a trusted CPE, boundary router 42 would also need to remark and police packets received at LP-I 110a.
Following classification (and marking in the case of traffic received at LP-2110b), traffic is forwarded to an appropriate physical port 116 or logical port 110 by forwarding function 118.
In contrast to edge router 34 of Figure 5, which utilizes classifiers to perform the full forwarding lookup, boundary router 42 employs an alternative design in which forwarding function 118 accesses forwarding table 128 with a packet's DA to determine the output port, namely, LP-1 110a, LP-2110b, or PHA-1 116a in this example. In the case of a non-VPN
router, forwarding table 128 is populated by generic IP routing protocols (e.g., Border Gateway Protocol (BGP)) or static configuration (e.g., association of the 24-bit IP address prefix "d.e.f." with LP-2 110b).
An alternative implementation could centrally place . the IP lookup forwarding function in forwarding function 62. The exemplary implementation shown in Figure 6 assumes that boundary router 42 sends all traffic bound for the network core to only one of the physical ports 116 connected to a core router. In other embodiments, it is possible, of course, to load balance traffic across physical ports 116. In addition, implementations omitting the core router or employing one or more logical ports to one or more core routers are straightforward extensions of the depicted design.
For traffic communicated to access network 38 through boundary router 42, classifier 132 accesses classifier table 134 utilizing the DSCP of the packets to direct each packet to the appropriate one of queues QO-Q-2 for the QoS indicated by the packet's DSCP.
For a customer that has purchased a Diffserv-enabled logical port 110, this has the effect of delivering the desired QoS since the source CPE has policed and marked the flow with appropriate DSCP
value. Although a best-effort customer is capable of receiving higher quality traffic, preventing such a one-way differentiated service would require significant additional complexity in the classifier and include distribution of QoS information via routing protocols to every edge router in a service provider network.
With reference now to Figure 6B, there is depicted a more detailed block diagram of a QoS-aware VPN boundary router 40, which may be utilized to provide Diffserv-enabled and DoS-protected VPN service within the network architecture depicted in Figure 4. As shown, boundary router 40 includes,a plurality of physical ports 226 for connection to core routers of Diffserv-enabled IP VPN network 44, a plurality of Diffserv-enabled logical ports 224 coupled to an access network 38 by a forwarding function 220 for incoming packets and a scheduler 222 for outgoing packets, and a forwarding function 228 that forwards packets between logical ports 224 and physical ports 226.
Each Diffserv-enabled logical port 224 implemented on boundary muter 40 serves a respective one of a plurality of VPNs. For example, Diffsezw-enabled logical port LP-A 224a serves a customer site belonging to VPN A, which includes customer sites having the 24-bit IP
address prefixes "a.b.c." and "a.b.d." Similarly, Diffserv-enabled logical port LP-B 224b serves a customer site belonging to VPN B, which includes two customer sites having the 24-bit IP
address prefixes "b.c.d." and "b.c.e." Diffserv-enabled logical ports 224 do not serve sites belonging to best effort IP public network 46 since such traffic is routed to boundary roisters 42, as shown in Figure 4.
As further illustrated in Figure 6B, each core-facing physical port 226 of boundary roister 40 is logically partitioned into a plurality of sub-interfaces implemented as logical tunnels 240.
As will be appreciated by those skilled.in the art, a tunnel may be implemented utilizing any of a variety of techniques, including an IP-over-IP tunnel, a Generic Routing Encapsulation (GRE) tunnel, an IPsec operated in tunnel mode, a set of stacked Multi-Protocol Label Switching (MPLS) labels, a Layer 2 Tunneling Protocol (L2TP), or ~a null tunnel. Such tunnels can be distinguished from logical ports in that routing information for multiple VPNs can be associated with a tunnel in a nested manner. For example, in the Border Gateway Protocol (BGP)/MPLS
VPNs described in IETF RFC 2547 [E. Rosen et al., "BGP/MPLS VPNs" March 1999], the topmost MPLS label determines the destination boundary muter while the bottommost label determines the destination VPN.
In operation, a classifier 230 on each of Diffserv-enabled logical ports 224 classifies packets flowing from access network 38 through boundary roister 40 to the network core of Diffserv-enabled IP VPN network 44 in accordance with the packets' DSCP values by reference to a respective classifier table 232. As depicted, classifier tables 232a and 232b are accessed utilizing the DSCP as an index to determine the appropriate one of queues QO-Q2 on physical port PHY-1 226a for each packet. Packets received by physical ports 226 are similarly classified by a classifier 250 by reference to a classifier table 254 to determine an appropriate one of queues QO-Q2 for each packet on one of logical ports 224. After classification (and optional (re)marking as shown at LP-B 224b), forwarding function 228 switches packets between logical ports 224 and physical ports 226 by reference to VPN forwarding tables 234a-234n, which are each associated with a respective VPN. Thus, for example, VPN forwarding table 234a provides forwarding routes for VPN A, and VPN forwarding table 234b provides forwarding routes for VPN B.
VPN forwarding,tables 234 are accessed utilizing the source port and DA as indices. For example, in the exemplary network configuration represented in forwarding table 234a, traffic within VPN A addressed with a DA having a 24-bit IP address prefix of"a.b.d."
traverses TNL-1 240a, and traffic received at TNL-1 240b is directed to LP-A 224a. Similar routing between TNL-2 240b and LP-B 224b can be seen in VPN routing table 234b. As discussed above, VPN
forwarding tables 234 can be populated by static configuration or dynamically utilizing a routing protocol.
Following processing by forwarding function 178, packets are each directed to the output port queue corresponding to their DSCP values. For example, packets marked with the QoS
class associated with DSCP 101 are placed in Q2, packets marked with the QoS
class associated with DSCP O10 are placed in Q 1, and traffic marked with DSCP 000 is placed in Q0. Schedulers 236 and 252 then schedule output of packets from queues QO-Q2 to achieve the requested QoS.
With reference now to Figure 7, there is illustrated an exemplary network architecture 150 that provides a network-based VPN solution to the DoS attack problem. In Figure 7, like reference numerals and traffic notations are utilized to identify features corresponding to features of network architecture 30 depicted in Figure 4.
As depicted, network architecture 150 of Figure 7, like network architecture 30 of Figure 4, includes a Diffserv-enabled IP VPN network 44, a best effort IP
public network 46, and a plurality of customer Local Area Networks (LANs) 32. As above, customer LANs 32a and 32b belong to the same VPN and each include one or more hosts 48 that can function as a transmitter and/or receiver of packets. Each customer LAN 32 is coupled by a CPE edge router 34 and a physical access link 153 to a respective access network (e.g., an L2 or L3 access network) 154. In contrast to access networks 38 of Figure 4, which have separate logical connections for QoS and best effort traffic, access networks 154 are only connected to boundary routers 156 of Diffserv-enabled IP VPN network 44, which have separate logical connections to boundary routers 42 of best effort IP public network 46. Thus, intra-VPN
traffic destined for network 44 and extra-VPN traffic destined for network 46 are both routed through boundary routers 156, meaning that work-conserving scheduling between the two classes of traffic is ..... . 1s.

advantageously permitted. However, as a consequence, the complexity of boundary routers 156 necessarily increases because each boundary router 156 must implement a separate forwarding table for each attached customer, as well as a full Internet forwarding table that can be shared among customers.
Referring now to Figure 8, there is depicted more detailed block diagram of a QoS-aware VPN boundary router in which the policers, shapers, schedulers, logical port access network connections and forwarding tables are configured to provide Diffserv-enabled and DoS-protected VPN service within the network architecture depicted in Figure 7. As shown, boundary router 156 includes a plurality of physical ports 176 for connection to network core routers, a plurality of Diffserv-enabled logical ports 174 coupled to access network 154 by a forwarding function 170 for incoming packets and a scheduler 172 for outgoing packets, and a forwarding function 178 that forwards packets between logical ports 174 and physical ports 176.
Because each CPE edge router 34 is coupled to a boundary router 156 by only a single access link through access network 154, each network customer site is served at boundaryrouter 156 by a pair of Diffserv-enabled logical ports 174, one for intra-VPN traffic and one for extra-VPN traffic. For example, Diffserv-enabled logical ports LP-Al 174a and LP-A2 174 serve a single customer site belonging to VPN A, which includes at least two customer sites having the 24-bit IP address prefixes "a.b.c." and "a.b.d." In the depicted embodiment, LP-A1 174a provides access to QoS traffic communicated across Diffserv-enabled IP VPN
network 44 to and from sites belonging to VPN A, while LP-A2174b provides access to best effort traffic to and from best effort IP public network 46.
As further illustrated in Figure 8, each core-facing physical port 176 of boundary router 156 is logically partitioned into a plurality of sub-interfaces implemented as logical tunnels 180.
As will be appreciated by those skilled in the art, a tunnel may be implemented utilizing any of a variety of techniques, including an IP-over-IP tunnel, a Generic Routing Encapsulation (GRE) tunnel, an IPsec operated in tunnel mode, a set of stacked Multi-Protocol Label Switching (MPLS) labels, or a null tunnel. Such tunnels can be distinguished from logical ports in that routing information for multiple VPNs can be associated with a tunnel in a nested manner. For example, in the Border Gateway Protocol (BGP)IMPLS VPNs described in IETF RFC
2547, the topmost MPLS label determines the destination boundary router while the bottommost label determines the destination VPN.
In operation, a classifier 182 on each of Diffserv-enabled logical ports 174 classifies packets flowing from access network 154 through boundary router 156 to the network core in accordance with the packets' DSCP values by reference to a respective classifier table 190. As depicted, classifier tables 190a and 190b are accessed utilizing the DSCP as an index to determine the appropriate one of queues QO-Q2 on physical port PHY-1 176a for each packet.
Packets received by physical ports 176 are similarly classified by a classifier 198 by reference to a classifier table 192 to determine an appropriate one of queues QO-Q2 for each packet on one of logical ports 174. After classification (and optional (re)maxking as shown at LP-A2 174b), forwarding function 178 switches packets between logical ports 174 and physical ports 176 by reference to VPN forwarding tables 194a-194n, which are each associated with a respective VPN and shared Internet forwarding table 195. Thus, for example, forwarding table 194a contains entries providing forwarding routes for VPN A, while Internet forwarding table 195 contains entries providing forwarding routes for packets specifying LP-A2 or TNL-2 (i.e., the logical interfaces configured for Internet access) as a source.
Forwarding tables 194 are accessed utilizing the source port and DA as indices. For example, in the exemplary network configuration represented in forwarding table 194a, intra-VPN traffic addressed with a DA having a 24-bit IP address prefix of"a.b.d."
traverses TNL-1 180a, while extra-VPN (i.e., Internet) traffic traverses TNL-2 180b (which could be a null tunnel). Forwarding table 194a further indicates that infra-VPN traffic received via TNL-1180a is directed to LP-A 1 174a, and all other traffic arriving from the Internet via tunnel TNL-2180b addressed with a DA having a 24-bit IP address prefix of"a.b.c." is sent to LP-A2174b. Traffic that terminates to other ports on boundary router 156 (i.e., traffic having a Local DA) is sent to other ports of boundary router 156 (indicated as LP-x). In other words, the entries in forwarding table 194a marked "Local" specify address prefixes other than those assigned to VPNs (e.g., a.b.c/24) that are assigned to interfaces on boundary router 156.
Following processing by forwarding function 178, packets are each directed to the output port queue corresponding to their DSCP values. For example, packets marked with the QoS
17 _.

class associated with DSCP 1 O1 are placed in Q2, packets marked with the QoS
class associated with DSCP 010 are placed in Q1, and best effort traffic marked with DSCP 000 is placed in Q0.
Schedulers 196 then schedule output of packets from queues QO-Q2 to achieve the requested QoS.
As has been described, the present invention provides an improved network architecture for providing QoS to infra-VPN traffic while protecting such flows against DoS
attack from sources outside the VPN. The present invention provides DoS-protected QoS to selected flows utilizing a network-based VPN service and a best effort Internet service connected to a CPE edge router using a L2 access network with appropriately configured routing protocols. Diffserv marking at the edge and handling in the network-based VPN core provides QoS to selected flows while logically partitioning infra-VPN and extra-VPN traffic to prevent DoS to a VPN network customer site due to traffic originating from outside of the customer's VPN
exceeding that site's access capacity. Even further protection from traffic originating from within the customer's VPN is possible using Intserv policy control, implemented on the CPE edge router and/or the QoS-aware boundary router, as described in IETF RFC 299 [Y. Bernet et al., "A
Framework for Integrated Services ~peration over Diffserv Networks" Nov. 2000].
The network architecture of the present invention may be realized in CPE-based and network-based implementations. The CPE-based implementation permits easy configuration of the access networks linking the CPE edge routers and service provider boundary routers and permits QoS to be offered to VPN sites without implementing Diffserv across the entire service provider network. The network-based configuration advantageously permits work conserving scheduling that permits extra-VPN traffic to utilize excess access capacity allocated to infra-VPN
traffic.
While various embodiments of the present invention have been described above, it should be understood that they have been presented by way of example only, and not limitation. Thus, the breadth and scope of the present invention should not be limited by any of the above-described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents. For example, although the present invention has been described with respect to preferred embodiments in which network-based VPNs are implemented within a i8 -19~-Diffserv network, it should be understood that the present invention is not restricted to use with Diffserv networks, but is instead to other network-based VPNs, which may be implemented, for example, utilizing BGP/MPLS as taught in RFC 2547 or virtual roisters as taught in RFC 2917 [I~. Muthukrishnan et al., "A Core MPLS IP VPN Architecture" Sept. 2000]. In addition, although Figures 3, 4 and 7 illustrate the connection of each CPE edge roister to a VPN network and a best effort network by one access link, it should be understood that, for redundancy, a CPE
edge roister may be connected by multiple access links to one or more access networks, which provide logical connections to one or more boundary roisters of each of the VPN and best effort networks. In such "dual homing" implementations, the multiple access links can be utilized in either a primary/backup or load-sharing arrangement through installation of static routes in the service provider boundary roisters or dynamic configuration of the service provider boundary roisters utilizing routing protocols (e.g., EBGP). This would require that the CPE edge roister implement multiple forwarding tables and separate instances of the routing protocol for the VPN
and Internet access address spaces. The implementation of such a CPE edge roister would be similar to that illustrated in Figure 8 and described in the associated text, with only a single VPN
table and a single table for Internet routes.

Claims

What is claimed is:

1. A network system, comprising:
a network infrastructure providing a virtual private network (VPN) and a best effort public network;
a first egress boundary router of said VPN and a second egress boundary router of said best effort public network that are each coupled for communication with an egress access network having an access link to which a destination host belonging to the VPN
is coupled;
a first ingress boundary router of the VPN and a second ingress boundary muter of the best effort public network, wherein said first ingress boundary muter transmits only packets originating from sources within the VPN and targeting the destination host to said first egress boundary router via said VPN, and wherein said second ingress boundary router transmits packets originating from sources outside the VPN and targeting the destination host to said second egress boundary router via said best effort public network;
wherein at least said first egress boundary router is configured to transmit packets received via said VPN and targeting said destination host onto the egress access network utilizing a separate logical connection than that employed for packets communicated over the best effort public network, such that the access link is protected from denial of service attacks originating from sources outside the VPN.

2. The network system of Claim 1, wherein at least said VPN is implemented within a Differentiated Services domain.

3. The network system of Claim 1, and further comprising:
the egress access network connected to at least said first egress boundary router; and an ingress access network connected to at least the first ingress boundary router.

4. The network system of Claim 3, wherein:
said ingress access network is connected to each of said first ingress boundary router and said second ingress boundary router;
said ingress access network has separate logical connections to said first and second ingress boundary routers for a customer premises equipment (CPE) edge muter;
and said ingress access network transmits packets having both source and destination addresses belonging to the VPN to said first ingress boundary router and transmits other packets to said second ingress boundary muter.

5. The network system of Claim 4, and further comprising a CPE edge router coupled to said ingress access network, wherein said CPE edge router includes a classifier that classifies at least some packets for routing to one of said first and second ingress boundary routers based at least in part on a host service markings in packet headers.

6. The network system of Claim 3, wherein:
said egress access network is connected to each of said first egress boundary router and said second egress boundary router;
said egress access network has separate logical connections to said first and second egress boundary routers for a customer premises equipment (CPE) edge router;
and said first egress boundary router transmits packets from the VPN to said CPE
edge router via a first of said logical connections and said second egress boundary router transmits packets from the best effort public network to said second ingress boundary router via a second of said logical connections.

7. The network system of Claim 6, wherein said egress access network assigns a higher priority to traffic received from said first egress boundary router than traffic received from said second egress boundary muter.

8. The network system of Claim 7, wherein said first egress boundary router shapes traffic destined for the destination host to prevent starvation of traffic of said second egress boundary router that is destined for the destination host.

9. The network system of Claim 6, wherein said first egress boundary router shapes traffic destined for the destination host to a first rate and said second egress boundary router shapes traffic destined the destination host to a second rate, wherein the sum of the first and second rates is no greater than a transmission capacity of said access link.

10. The network system of Claim 1, wherein said first ingress router includes:
first and second logical input interfaces for receiving traffic destined for the VPN and for the best effort public network, respectively;
first and second logical output interfaces for transmitting traffic over the VPN and the best effort public network, respectively; and a forwarding function that switches packets received at said first logical input interface to said first logical output interface and that switches packets received at said second logical input interface to said second logical output interface.

11. The network system of Claim 10, wherein said first egress router includes:
first and second logical input interfaces for receiving traffic from the VPN
and from the best effort public network, respectively;
a first logical output interface and a second logical output interface respectively coupled to separate first and second logical connections on said egress access network, wherein said first logical output interface transmits traffic received from the VPN utilizing said first logical connection and said second logical output interface transmits traffic received from the best effort public network utilizing the second logical connection; and a forwarding function that switches packets received at said first logical input interface to said first logical output interface and that switches packets received at said second logical input interface to said second logical output interface.

12. The network system of Claim 11, wherein said first egress boundary muter includes a scheduler, coupled to each of said first and second logical output interfaces, that transmits packets from said first and second logical output interfaces onto said egress access network, wherein said scheduler grants a higher priority to traffic from said first logical output interface than to traffic from said second logical output interface.

13. The network system of Claim 12, wherein said scheduler performs work-conserving scheduling on outgoing traffic from said first and second logical output interfaces.

14. The network system of Claim 11, wherein the VPN is one of a plurality of VPNs, and wherein said forwarding function has a corresponding plurality of VPN
forwarding tables and a shared forwarding table for best effort traffic.